Sentiment Analysis Flashcards
Sentiment Analysis
Extracting opinions from text
Document sentiment analysis
Detects whether a document is positive, negative, or neutral
Feature-based sentiment analysis
Identifies sentiment towards particular features of an object
Opinion spam detection
Automatically distinguish genuine reviews from fake reviews
Opinion lexicon
Positive and negative words and phrases
Can be automatically learned using a dictionary
Can be learned from text
Term AND Term have same polarity
Term BUT Term have opposite polarity
Document sentiment analysis as supervised machine learning
Use known opinion words and phrases
Keep track of negation
Bag of words and word n-grams
Need domain-specific documents and word lists to deal with ambiguity
Document sentiment analysis as unsupervised machine learning
Identify pairs of adjectives and one preceding/succeeding word
Classify word pairs as positive or negative by measuring the association with the terms “excellent” and “poor”
Pointwise mutual information
Used to measure the association between two items A and B
Opinion spam detection as a supervised machine learning problem
Features are the words occurring in the text
- Register
- Number of typos
- How specific (concrete vs vague)
- How much do they talk about themselves
